Arthur

3/3 Not out yet, but on my dock every day. Hear first when it ships: aaa.schmidtpabst.com#/portal/signup #macOS #indiedev #macapp #buildinpublic

2/3 Unplug a drive and its files grey out instead of vanishing. Search finds text inside images, on the Mac. Or click a color swatch and see every image that has it. No account, no cloud, no telemetry. #macdev #macapp #buildinpublic

1/3 My files live everywhere: Photoshop on one SSD, photos on three SD cards, PDFs in a folder I forgot existed. So I built Caracara, a Mac app that gathers everything for a project without moving, copying or duplicating a single file. #buildinpublic #macapp #macdev

Too much in your head, no clear place to start? I’m building One Knot: an app for speaking or typing tangled thoughts and finding one next step. ADHD is central to its design; feeling overwhelmed is the starting point. What would you want help untangling? #buildinpublic #iphoneapp #adhd

FOSS colleagues: Blueprint Studio is MIT-licensed. Inspect the rules, adapt the tool, or self-host it. What would you change for your team? Source: github.com/SirArthur23/... Try it: blueprintstudio.schmidtpabst.com #FOSS #buildinpublic #aigouvernance

Sunday has a side quest. The quest giver is a website. Spiegel des Universums is a street art game: get a two-word glitch phrase, a rando destination, go there, and see what the phrase becomes in that place. Photograph the result, share your trace on the website. #UrbanPlay #streetart #ritual

I built Dispatch to make my own social media routine easier: campaign copy, dates, and publishing progress in one Mac app. I'd like to share it with you, free, before the end of 2026. Explore my work: aaa.schmidtpabst.com #buildinpublic #FOSS #solodev #macapp

A memo needs someone with a reason to write it. The Veil's Inference Engine is being built to reconstruct documents through chosen lenses. Paid-subscriber access comes when ready. Join the research. Subscribe: theveil.schmidtpabst.com #worldbuilding #scifi #participatoryart

Der Ausstiegstest für Ihre Software: Können Sie einen Vorgang exportieren und anderswo weiterbearbeiten — samt Kommentaren und Verknüpfungen? Urteilskraft 002: Können Sie Ihren Anbieter verlassen? schmidtpabst.com/de/blog/urte... #DigitaleSouveränität #German #KI

I’m building One Knot, a coming thought-sorting app from AAA. A six-week introduction starts here and on LinkedIn on Monday, 26 October. Meet the app: aaa.schmidtpabst.com/news/one-knot #buildinpublic #indydev #iOS #ai

I've opened Blueprint Studio's public beta: a free, MIT-licensed tool for planning AI-assisted work with human owners and review points. I'm Arthur, an AI trainer, consultant, and designer. Bring one use case and try it: blueprintstudio.schmidtpabst.com #augmentedproductdesigner #ai

I gave one app four names, almost shipped a "Salad Bundle", and ended up with an aviary. Naming apps in 2026: 40 bird names checked, 2 survivors. Story of Caracara, Pirol & Curlew: aaa.schmidtpabst.com/news/from-sa... #indiedev #macOS #buildinpublic

The Institute invites you into The Veil: sci-fi told through damaged evidence. Our Inference Engine is under development for future paid-subscriber access. The archive is open. Join the research. Subscribe: theveil.schmidtpabst.com

The Workbench is not the embarrassing room before the Window. It is where the promises are tested. “Not ready yet” can be evidence of product judgment. aaa.schmidtpabst.com

Scattered AI use becomes shared practice when a team can compare: tasks, inputs, boundaries, evidence, review, approval, escalation, ownership, and what it learned when the workflow failed.

Your team may not need more AI experiments. It may need a shared way to learn from them: boundaries, evidence, review, approval, escalation, ownership. HITM Labs turn scattered use into practice people can improve together. ai.schmidtpabst.com/contact/

Four ways into my work: Learn — AI Lab or HITM Labs Decide — consultation and governance Design — idea to working prototype Follow — AAA builds and releases Bring the real situation. I will help find the right door. schmidtpabst.com/contact/ #AugmentedProductDesigner #AITraining

Built with AI. Runs without it. AI helped me implement the apps. The apps do not need to send working data to a remote model. Local-first is not magical security. It is a deliberately smaller attack surface. aaa.schmidtpabst.com #LocalFirst #PrivacyByDesign #ResponsibleAI #IndieSoftware

When generation becomes cheap, selection becomes more valuable. NC Tarot was not a task of making a few striking images. It was holding one system across 78 cards, a handbook, and a 400-page course. Constraint, rejection, selection, integration. design.schmidtpabst.com/work/

Useful AI literacy should increase capability and skepticism at the same time. Safe input. Missing information. Verification. Human-owned decisions. A credible option to stop. Use it without trusting it. ai.schmidtpabst.com/learn/ #AILiteracy #ResponsibleAI #CriticalThinking #AITutorials

Some ideas do not need another presentation. They need to become something people can use. I work from product framing through flows, interface, visual system, AI-assisted implementation, testing, and working prototype. design.schmidtpabst.com/contact/

AI in my apps: code, refactoring, debugging, tests, API explanations, implementation speed. Arthur: product definition, flows, interface, constraints, evaluation, rejection, testing, release, support, responsibility. Assistance is substantial. Authority still has a location.

Panic Defense is a design study for a person who cannot give the interface full attention. Minimal reading. Motor tolerance. Clear choice. No engagement loop. A clean exit. Human control changes when the human is under pressure. design.schmidtpabst.com/panic-defense/ #UXDesign #HumanCenteredDesign

Arthur’s AI Lab has three tracks: Green: Everyday AI Yellow: Systems & Agents Purple: Strange Experiments Not tiers. Three different reasons to learn in public. The first pilots are in production and stay free. ai.schmidtpabst.com/learn/

Building software changed how I teach AI. Teaching changed how I build software. Products expose failure and maintenance. Students expose assumptions and unclear explanations. That is why “augmented product designer” and “practical AI tutor” belong together for me. #AugmentedProductDesigner

Applied Lab and Quest Lab teach the same HITM method. Applied: sober, direct, implementation-led. Quest: playful, lower-fear, socially energizing. Different choreography. Same standard of judgment. ai.schmidtpabst.com/labs/ #AITraining #LearningDesign #HumanInTheMiddle #TeamDevelopment

cucumber handles budgets, bills, incoming money, and final invoices. It will stay on the Workbench until calculations, exports, backups, recovery, and financial boundaries survive appropriate suspicion. Money dislikes improvisation. aaa.schmidtpabst.com/apps/cucumber/

A portfolio shows the result after uncertainty has been edited out. A public workbench can show what works, what failed, what remains uncertain, and what happens next. For AI-assisted products, maturity is part of the product. schmidtpabst.com/projects/ #BuildInPublic #AugmentedProductDesigner

card&board now plans the work involved in building card&board. That is either a convincing acceptance test or how software becomes self-aware ;) For now: private working app, used daily, still not ready for sale. aaa.schmidtpabst.com/news/card-bo... #IndieSoftware #ProductDesign #BuildInPublic

I am pro-AI and critical of the power, labor, resource, and political structures gathering around it. Not contradictory. The technology can expand who gets to build. It can also concentrate authority. That tension is where the design work begins. #ResponsibleAI #AIEthics #DesignThinking

Human-in-the-loop adds a reviewer to the output. Human in the Middle designs human authority into the workflow: framing, boundaries, evidence, approval, escalation, and stopping. The difference is architectural. ai.schmidtpabst.com/human-in-the... #HumanInTheMiddle #AIGovernance #ResponsibleAI

A human decision gate should appear where the consequence occurs. Not at the end, when the output feels finished and approval is the path of least resistance. GLASSBOX is my design study for making that choice real. design.schmidtpabst.com/glassbox/

A misclick exposed a missing recovery layer in Aerial View. Now the app backs up its database whenever it exits normally. AI helped implement the repair. Daily use supplied the requirement. aaa.schmidtpabst.com/news/aerial-...

An empty approval gate is worse than no gate. It manufactures evidence of diligence. I shipped one: "Data boundary approved" — a checkbox with no boundary field behind it. What I learned building three products: schmidtpabst.com/judgment-is-... #AI

Someone printing a protest poster at 2am knows things your software doesn't. So Boar's four guards — overflow, contrast, resolution, safe area — explain the problem and offer a fix, but never block. Any of them can be waved off. schmidtpabst.com/boar #ProductDesign

IO: localhost is a tiny Mac app for seeing and controlling local servers in one place. It will be free, open source, signed, notarized, and documented. Still on the Workbench until those promises are actually true. aaa.schmidtpabst.com/apps/io-loca... #AIProductUX #HumanCenteredAI #ResponsibleAI

If an AI system reveals its plan only after acting, the “human checkpoint” is mostly theatre. Visible intent means showing the right information while a person can still change the direction. design.schmidtpabst.com

One practice, four jobs: Code:Emotion supplies design judgment. AAA supplies working proof. HITM names the method. The Labs make it teachable. Practice makes teaching credible. Teaching makes practice transferable. schmidtpabst.com

Four small Mac apps, each with one responsibility: IO: localhost — local servers Aerial View — material card&board — work cucumber — money Useful alone. Cooperative when needed. No giant dashboard committee. aaa.schmidtpabst.com

Most AI products demonstrate what the model can do. The harder design question is what the human needs in order to understand, approve, correct, stop, and own the result. That is the human-control layer. design.schmidtpabst.com

“Can AI do this?” is a capability question. “Should we automate it, and under what conditions?” is a responsibility question. Most organizations need the second question more urgently than another tool list. ai.schmidtpabst.com

Working definition: An augmented product designer uses AI to expand how much of a product they can conceive, prototype, implement, test, and maintain; without giving the machine product authority. More range. Same ownership. Where would you draw the boundary? #ProductThinking #DesignProcess

One of the nicer AI stories: An experienced product designer gained enough implementation leverage to build his own native apps. AI wrote plenty of code. It did not decide what the products should be—or whether they were ready. aaa.schmidtpabst.com #IndieSoftware #ProductDesign

I do not want to teach AI only from slides. I want to teach from products with bugs, backups, edge cases, release gates, and consequences. What failed in practice is often the most useful part of the lesson. ai.schmidtpabst.com #AITeaching #ResponsibleAI #HumanInTheMiddle #LearningDesign

AI does not make thirty years of design judgment obsolete. It creates a harder design brief: Can a person see, check, stop, modify, and reverse what the system is doing? That human-control layer is product design. design.schmidtpabst.com

I did not become an “AI designer.” I became an augmented product designer. More implementation range. More ability to move from idea to working software. Not less responsibility. What should someone using that title be expected to own? schmidtpabst.com

The Veil is built as archaeology of the future. Not prophecy. Not a lore wiki. Fragments arrive damaged. Every interpretation has pressure, bias, and uncertainty. Start reading: theveil.schmidtpabst.com

The hub is the map. The branches are where the work happens: Design: design.schmidtpabst.com AI consultation: ai.schmidtpabst.com Root: schmidtpabst.com

Some briefs are too weird for a normal agency menu. That is often where the useful design question starts. What structure does this need, so the weirdness becomes legible? design.schmidtpabst.com